4. Multi-Degree-of-Freedom and Continuous Systems

Introduction and Context

Many mechanical systems cannot be represented adequately by a single displacement coordinate. A vehicle body may bounce, pitch, and roll; a machine foundation may translate and rotate; a rotor may undergo lateral bending and torsional motion; and a structure may possess thousands of interacting deformation coordinates. Such systems are described as multi-degree-of-freedom (MDOF) systems. When the number of coordinates becomes very large, or theoretically infinite, the system is treated as a continuous system.

The transition from single-degree-of-freedom analysis to MDOF and continuous-system analysis is fundamental in mechanical engineering. It introduces coupling, mode shapes, orthogonality, modal coordinates, matrix eigenvalue problems, and distributed-parameter models. These concepts provide the basis for finite-element vibration analysis, experimental modal testing, rotor dynamics, structural health monitoring, and active vibration control.

The central objective is not merely to calculate frequencies. A useful dynamic model must identify how inertia, energy dissipation, and elastic forces are distributed and coupled. The resulting model supports decisions about resonance avoidance, isolation, fatigue, controller design, sensor placement, and structural modification.

Learning Objectives

After studying this chapter, the reader should be able to:

1. Formulate mass, damping, and stiffness matrices for coupled mechanical systems.

2. Derive and solve the eigenvalue problem for natural frequencies and mode shapes.

3. Explain modal orthogonality, normalization, and modal participation.

4. Transform physical coordinates into modal coordinates.

5. Apply modal truncation and other model-reduction techniques responsibly.

6. Formulate continuous-system models for shafts, beams, plates, and coupled structures.

7. Interpret boundary conditions, mode shapes, resonance, and damping in distributed systems.

8. Select suitable analytical, numerical, and experimental methods for engineering applications.

From Coupled Coordinates to Matrix Equations

Consider an undamped linear system with n generalized coordinates collected in the vector

q(t) = [q1 q2 ... qn]ᵀ.

Using Lagrange’s equations, the equations of motion can be written as

M q̈ + Kq = f(t),

where M is the mass matrix, K is the stiffness matrix, and f(t) is the external-force vector. With viscous damping included,

M q̈ + C q̇ + Kq = f(t).

The matrices have clear physical meanings. M relates acceleration to inertial force, C relates velocity to dissipative force, and K relates displacement to restoring force. For a conservative mechanical system, M and K are usually symmetric. A physically meaningful mass matrix is positive definite, while a stiffness matrix is positive definite when rigid-body modes are absent.

For a two-mass system connected by springs, let two masses m1 and m2 be connected to fixed supports by springs k1 and k2, with a coupling spring kc between them. The equations are

m1 ẍ1 + (k1+kc)x1 − kcx2 = f1(t),

m2 ẍ2 + (k2+kc)x2 − kcx1 = f2(t).

Thus,

M = [[m1,0],[0,m2]],

K = [[k1+kc,−kc],[−kc,k2+kc]].

The negative off-diagonal terms indicate that a displacement of one mass produces a force opposing relative displacement between the masses. Matrix entries should therefore be interpreted physically rather than treated as merely algebraic coefficients.

A consistent finite-element formulation often produces off-diagonal mass terms as well. These terms represent inertial coupling associated with the assumed interpolation functions. Concentrated-mass and consistent-mass models can yield slightly different frequencies, particularly for higher modes.

Free Vibration and the Eigenvalue Problem

For free, undamped vibration, set f(t)=0 and seek a harmonic solution

q(t)=φ sin(ωt),

where φ is a constant mode-shape vector and ω is the corresponding natural frequency. Substitution gives

(K−ω²M)φ=0.

A nontrivial solution exists only when

det(K−ω²M)=0.

This is the generalized eigenvalue problem. Its eigenvalues are ωr², and the positive square roots ωr are the natural frequencies. The associated eigenvectors φr are the mode shapes.

For an n-degree-of-freedom system, there are generally n modes. The first mode has the lowest frequency and usually represents the smoothest deformation; higher modes contain progressively more spatial variation. This interpretation is useful but not universal: local components, mass distributions, and constraints can cause higher modes to dominate a particular measurement or excitation.

For the two-mass system, the characteristic equation is

det(K−ω²M)=0.

Solving it produces two natural frequencies and two corresponding mode shapes. The relative signs in a mode shape are important. If both components have the same sign, the masses move broadly in phase; if their signs differ, they move out of phase. The scale of an eigenvector is arbitrary, but its component ratios and signs are physically meaningful.

Orthogonality and Normalization

For symmetric M and K, distinct modes satisfy

φrᵀMφs=0, r≠s,

and

φrᵀKφs=0, r≠s.

These are mass and stiffness orthogonality relations. They allow a coupled system to be decomposed into independent modal equations. A common normalization is mass normalization:

φrᵀMφr=1.

Under this convention, φrᵀKφr=ωr². Another practice is to set one component equal to unity, which is convenient for plotting but does not produce unit modal mass. Confusing these normalization conventions can lead to incorrect modal-force and response calculations.

Rigid-body modes require special attention. If a structure is free to translate or rotate, one or more natural frequencies are zero. The stiffness matrix is then positive semidefinite rather than positive definite. Rigid-body modes are not errors; they represent motions that do not generate elastic restoring force.

Modal Transformation and Dynamic Decoupling

Let the modal matrix be

Φ=[φ1 φ2 ... φn].

Introduce modal coordinates η through

q=Φη.

Substitution into the physical equations and premultiplication by Φᵀ gives

ΦᵀMΦ η̈ + ΦᵀCΦ η̇ + ΦᵀKΦ η = Φᵀf(t).

For a classically damped system, the transformed matrices are diagonal:

Mη̈ + Cη̇ + Kη = f(t).

With mass-normalized modes, M=I and K=diag(ω1²,...,ωn²). Each modal coordinate then satisfies an independent single-degree-of-freedom equation,

η̈r + 2ζrωrη̇r + ωr²ηr = φrᵀf(t),

where ζr is the modal damping ratio.

The modal force depends on the spatial distribution of excitation. A force applied at a node where a particular mode has a small amplitude will weakly excite that mode. Conversely, excitation near an antinode can produce a large response. This explains why a structure may resonate strongly at one location but show little motion at another.

Damping requires caution. Proportional, or classical, damping is commonly represented by

C=αM+βK.

This assumption preserves modal decoupling and gives

ζr=(α+βωr²)/(2ωr).

Real joints, friction interfaces, viscoelastic materials, and fluid interactions often produce nonproportional damping. In that situation, real undamped modes do not fully decouple the damped equations, and complex eigenvectors or state-space methods may be required.

Harmonic Response and Resonance in MDOF Systems

For harmonic force f(t)=Re{F e^{iΩt}}, the steady-state response is written q(t)=Re{Qe^{iΩt}}, where

Q=[−Ω²M+iΩC+K]⁻¹F.

The matrix in brackets is the dynamic stiffness matrix. Its inverse is the frequency-response function (FRF) matrix. The element Qj/Fi describes the response at coordinate j due to excitation at coordinate i.

In modal form, the response can be expressed approximately as a sum of modal contributions. Each mode contributes strongly when the forcing frequency is near its natural frequency, subject to damping and modal participation. Antiresonances can occur because modal contributions cancel at a particular response coordinate. Therefore, a frequency-response curve should not be interpreted only by counting peaks; zeros, phase changes, and measurement location also provide important structural information.

For base excitation, the forcing is not always represented as an ordinary applied force. A coordinate transformation or an equivalent inertial-force formulation may be necessary. Errors in the excitation model are a frequent source of incorrect vehicle and machinery vibration predictions.

Model Reduction and Modal Truncation

A detailed finite-element model may contain millions of degrees of freedom, while the frequency range of interest may involve only a small number of modes. Model reduction seeks a smaller model that retains the dominant dynamic behavior.

The simplest approach is modal truncation. Retain the modes within, or sufficiently influential to, the frequency range of interest:

q≈Φrηr,

where Φr contains only selected modes. This is effective for lightly damped systems and low-frequency response, but omitted modes can affect static flexibility, high-frequency response, local stresses, and control spillover.

Static correction methods, such as residual-flexibility or constraint-mode approaches, improve the representation of the influence of omitted high-frequency modes. Component mode synthesis, including fixed-interface methods, divides a large structure into components and combines internal normal modes with interface constraint modes. This is valuable for assemblies such as engines, gearboxes, and vehicle bodies.

Reduction should be validated by comparing retained-model predictions with a reference model over the intended frequency and operating range. A reduced model that reproduces the first few frequencies but gives incorrect interface forces is not adequate for component integration or controller design.

Continuous Systems: From Partial Differential Equations to Modes

A continuous body possesses an effectively infinite number of degrees of freedom. Its displacement depends on both position and time, such as u(x,t) for a shaft or beam, and w(x,y,t) for a plate. The governing equations are partial differential equations (PDEs), accompanied by boundary and initial conditions.

A common solution assumes separation of variables:

u(x,t)=X(x)T(t).

This yields a spatial eigenvalue problem and a temporal harmonic equation. The admissible spatial functions are mode shapes, and their associated frequencies are natural frequencies. Boundary conditions determine the allowable functions and therefore the spectrum.

The key distinction between discrete and continuous systems is not that one has modes and the other does not. Both possess modes. The distinction is that a continuous system has infinitely many modes, while a finite MDOF model has a finite number. Finite-element and assumed-modes methods approximate the continuous system using a finite basis.

Longitudinal and Torsional Vibration of Shafts

For a uniform shaft undergoing longitudinal motion u(x,t), neglecting damping,

∂²u/∂t² = c² ∂²u/∂x²,

where c=(E/ρ)¹ᐟ² is the longitudinal wave speed. For torsional motion θ(x,t),

∂²θ/∂t² = c_t² ∂²θ/∂x²,

where c_t=(G/ρ)¹ᐟ² for a uniform shaft under the elementary torsional model.

Boundary conditions determine the modes. A fixed end has zero displacement or rotation. A free end has zero axial force or torque, which corresponds to a zero spatial derivative in the ideal uniform model. A shaft fixed at both ends has sinusoidal modes with no rigid-body torsional mode; a freely suspended shaft has a rigid-body mode at zero frequency.

In practical rotating machinery, shafts may also experience bending, gyroscopic effects, bearing flexibility, disk inertia, and internal damping. The resulting rotor model may be nonsymmetric, and forward and backward whirl modes can have different frequencies.

Transverse Vibration of Beams

For an Euler–Bernoulli beam with transverse deflection w(x,t),

ρA ∂²w/∂t² + EI ∂⁴w/∂x⁴ = 0.

Here E is Young’s modulus, I is the second moment of area, A is cross-sectional area, and ρ is density. The fourth-order spatial derivative means that four boundary conditions are required. At a clamped end, w=0 and ∂w/∂x=0. At a free end, bending moment and shear force vanish, giving ∂²w/∂x²=0 and ∂³w/∂x³=0 for a uniform beam.

Euler–Bernoulli theory neglects rotary inertia and shear deformation. It is generally suitable for slender beams at lower frequencies. Timoshenko beam theory includes both effects and becomes more appropriate for short, thick beams or higher-frequency analysis. Selecting Euler–Bernoulli theory for a flexible, stocky component can produce inaccurate frequencies and mode shapes.

For a simply supported uniform beam, the mode shapes are sinusoidal and the natural frequencies increase approximately with the square of the mode number. For other boundary conditions, the spatial functions involve combinations of trigonometric and hyperbolic functions. Numerical methods are often preferable for stepped, curved, or elastically supported beams.

Plates and Shell-Like Components

For a thin, isotropic Kirchhoff plate, the transverse vibration equation can be represented as

D∇⁴w + ρh ∂²w/∂t² = 0,

where D=Eh³/[12(1−ν²)] is the flexural rigidity, h is thickness, and ν is Poisson’s ratio. Plate modes vary in two spatial directions and may be classified by nodal lines. A mode with many nodal lines has higher spatial complexity and generally higher frequency.

Thick plates require shear deformation and rotary inertia corrections. Curvature, anisotropy, stiffeners, cut-outs, and attachments usually prevent simple closed-form solutions. Finite-element analysis, spectral methods, or experimental modal analysis may then be used.

Coupled Structures and Interface Dynamics

Real engineering assemblies combine components with different masses, stiffnesses, materials, and damping mechanisms. Coupling may occur through bolts, welds, bearings, seals, flexible mounts, fluid films, or contact interfaces. The interface can transmit both forces and moments, and its properties may depend on preload, temperature, frequency, and amplitude.

A useful assembly procedure is to formulate component equations, define compatible interface coordinates, and impose equilibrium of interface forces. In finite-element practice, assembly adds elemental matrices into global matrices according to connectivity. Incorrect node numbering, duplicated interface degrees of freedom, or inconsistent coordinate systems can generate spurious modes.

Substructuring and component mode synthesis permit a large assembly to be studied using reduced component models. This supports design iteration because a component can be changed without rebuilding the entire model. However, reduction must preserve the interface behavior relevant to the intended analysis.

Experimental and Computational Modal Analysis

Computational modal analysis obtains frequencies and modes from a mathematical model. Experimental modal analysis estimates them from measured excitation and response, commonly using impact hammers or electrodynamic shakers. Accelerometers, laser vibrometers, and strain sensors provide response measurements.

Important experimental considerations include excitation bandwidth, sensor mass, boundary-condition realism, leakage, noise, and adequate spatial sampling. A mode can be missed if sensors are placed near nodal points. Frequency-response functions should be checked for coherence, phase consistency, and repeatability.

Model validation compares measured and predicted frequencies, mode shapes, and FRFs. Frequency agreement alone is insufficient because distinct mode shapes can possess similar frequencies. Correlation measures such as the modal assurance criterion compare the alignment of two mode vectors. Discrepancies should be interpreted physically: they may arise from uncertain joints, incorrect material properties, unmodeled damping, or unrealistic support conditions.

Applied Case: Vibration Reduction in a Motor–Pump Skid

Background

The following is an illustrative engineering case. A motor drives a centrifugal pump mounted on a steel skid. During operation, operators observe excessive vibration near the pump bearing at a speed close to one of the predicted structural resonances.

Problem Definition

A preliminary finite-element model includes the motor, pump, skid, and foundation mounts. The first vertical mode is predicted near the operating speed, but the measured vibration peak is shifted and is larger than expected. The engineering team must determine whether the problem is caused by rotor excitation, support flexibility, structural coupling, or modeling error.

Analysis

The assembly is represented by

M q̈+C q̇+Kq=f(t).

The model includes translational and rotational interface coordinates at the mounts. An initial eigenanalysis identifies a vertical bending mode involving the pump and skid, while the motor moves with a smaller amplitude. Experimental impact testing shows a similar mode shape but a lower natural frequency than predicted. This indicates that the effective mount stiffness or joint stiffness is lower than assumed.

The measured FRF also shows a strong response at the pump bearing because that location lies near an antinode of the structural mode. A sensor on the motor casing records less vibration, not because the excitation is absent, but because the motor is near a modal node for that mode.

Engineering Response

The team updates the model using measured mount stiffness and incorporates a realistic damping estimate. A modal reduction retains the first several skid modes and includes residual flexibility so that the pump–foundation interface remains accurate. Three design alternatives are evaluated: increasing skid stiffness, changing mount stiffness, and shifting operating speed.

Increasing skid stiffness raises the resonance but transfers more force to the foundation. Changing mount stiffness shifts the coupled mode and may reduce force transmission while increasing static displacement. The selected solution combines a modest skid modification with revised mount properties and an operating-speed restriction during transient passage through resonance.

Outcome and Learning Points

The revised model reproduces the measured modal frequencies and the dominant FRF peak with acceptable engineering accuracy. The vibration level decreases because the operating point is separated from the coupled structural mode and the pump-bearing response is reduced.

The case illustrates several principles: modes belong to the assembled structure, not necessarily to one component; a local measurement may exaggerate or conceal a global mode; joint and mount properties often control prediction quality; and model reduction is useful only when interface dynamics are preserved.

Limitations, Risks, and Common Errors

Several mistakes recur in MDOF and continuous-system analysis:

• treating a coupled system as independent single-degree-of-freedom units;
• using incorrect or inconsistent units in M, C, and K;
• ignoring rigid-body modes or accidentally constraining them;
• assuming proportional damping without checking whether the physical damping mechanism supports that assumption;
• confusing eigenvector scaling with physical amplitude;
• applying inappropriate beam or plate theory;
• neglecting joint, mount, bearing, or contact stiffness;
• interpreting every frequency-response peak as a structural natural frequency;
• truncating modes without considering residual flexibility or control spillover;
• relying on frequency agreement while ignoring mode-shape correlation;
• using insufficient sensor density in experimental modal testing.

Linear models also have limits. Large-amplitude motion, friction, backlash, contact loss, material yielding, fluid–structure interaction, and temperature-dependent properties may introduce nonlinear or time-varying behavior. In such cases, linear modes remain useful for initial diagnosis, but nonlinear simulation or operating-condition testing may be necessary.

Professional and Emerging Practice

Modern vibration analysis increasingly combines finite-element models, test data, operational measurements, and data-driven diagnostics. Digital twins may use reduced-order structural models updated by sensor data. Automated parameter estimation can identify stiffness and damping changes, while machine-learning methods can classify operating states or detect deviations from a baseline.

These technologies do not remove the need for mechanical judgment. Poor sensor placement, biased training data, uncertain boundary conditions, and unvalidated models can produce confident but incorrect conclusions. Engineering governance requires traceable assumptions, validation against measurements, version control of models, and clear communication of uncertainty. Safety-critical decisions should not rely on an opaque algorithm without physical verification.

Best practice is to begin with a physically interpretable model, establish boundary conditions carefully, perform convergence studies, compare predictions with measurements, and document the frequency range and assumptions for which the model is valid. For design decisions, sensitivity analysis is especially valuable: it identifies whether mass, stiffness, damping, joint properties, or operating speed offers the most effective route to vibration reduction.

Chapter Summary

Multi-degree-of-freedom systems are represented by matrix equations involving mass, damping, and stiffness matrices. Free vibration leads to the generalized eigenvalue problem, whose eigenvalues provide natural frequencies and whose eigenvectors provide mode shapes. Symmetric systems possess mass and stiffness orthogonality, allowing modal transformation and, under classical damping, decoupling into independent modal equations.

Harmonic response is obtained through the dynamic-stiffness matrix or modal superposition. Modal participation depends on excitation and measurement locations, while antiresonance results from modal cancellation. Model reduction, modal truncation, residual flexibility, and component mode synthesis make large structural models computationally manageable, but they must preserve the dynamics relevant to the application.

Continuous systems possess infinitely many modes and are governed by partial differential equations with boundary conditions. Shafts, beams, plates, and coupled structures require appropriate theories and careful treatment of supports, interfaces, damping, and material behavior. Computational predictions should be validated through experimental modal analysis and response measurements. Sound engineering practice combines mathematical rigor, physical interpretation, model validation, and explicit recognition of uncertainty.

Key Terms

Mass matrix: Matrix representing inertial relationships among generalized coordinates.

Damping matrix: Matrix relating generalized velocities to dissipative forces.

Stiffness matrix: Matrix relating generalized displacements to elastic restoring forces.

Generalized eigenvalue problem: Equation (K−ω²M)φ=0 used to obtain natural frequencies and modes.

Natural frequency: Frequency of free vibration of a system in a particular mode.

Mode shape: Spatial pattern of relative motion associated with a natural frequency.

Modal matrix: Matrix whose columns are selected mode-shape vectors.

Modal coordinate: Coordinate describing the amplitude of a particular mode.

Mass orthogonality: Vanishing mass inner product between distinct modes.

Modal normalization: Scaling of a mode shape according to a specified convention.

Modal participation: Degree to which a force or motion excites a particular mode.

Dynamic stiffness: Frequency-dependent matrix K−Ω²M+iΩC.

Frequency-response function: Complex response-to-input relationship as a function of frequency.

Modal truncation: Approximation obtained by retaining only selected modes.

Residual flexibility: Static or quasi-static contribution of omitted higher modes.

Continuous system: System with distributed properties and theoretically infinite degrees of freedom.

Boundary condition: Physical constraint governing displacement, slope, force, or moment at a boundary.

Euler–Bernoulli beam: Beam model neglecting shear deformation and rotary inertia.

Component mode synthesis: Substructuring method combining reduced component models.

Modal assurance criterion: Correlation measure used to compare mode shapes.

Review Questions

Short Answer Questions

1. What physical information is represented by the off-diagonal terms of a stiffness matrix?

2. State the generalized eigenvalue equation for an undamped MDOF system.

3. What is meant by mass normalization of a mode shape?

4. Why can a free structure possess a zero natural frequency?

5. Distinguish between a discrete MDOF system and a continuous system.

Long Answer Questions

1. Derive the modal transformation for an undamped MDOF system and explain how orthogonality produces decoupled equations.

2. Discuss the formulation and assumptions of Euler–Bernoulli beam vibration theory, including the role of boundary conditions.

3. Explain harmonic response analysis using the dynamic-stiffness matrix and modal superposition.

4. Compare modal truncation, residual-flexibility methods, and component mode synthesis.

5. Explain why damping may prevent modal decoupling and describe the significance of proportional damping.

Analytical / Application Questions

1. A two-mass system has unequal masses and a coupling spring. Explain how the signs and relative magnitudes of the components of each mode shape should be interpreted.

2. A finite-element model predicts the first natural frequency accurately but gives an incorrect pump-bearing response. Identify likely causes and propose a validation procedure.

3. A short, thick beam is analyzed using Euler–Bernoulli theory and shows disagreement with test data at higher frequencies. Decide whether Timoshenko theory is warranted and justify your decision.

22. Multiple Choice Questions with Answers

Question 1

For an undamped MDOF system, which equation determines the natural frequencies?

A. det(K+ωM)=0

B. det(K−ω²M)=0

C. det(M−ω²K)=0

D. det(C−ωK)=0

Correct Answer: B. det(K−ω²M)=0

Question 2

If two distinct modes of a symmetric system are mass-orthogonal, which relation holds?

A. φrᵀMφs=1

B. φrᵀKφs=ωrωs

C. φrᵀMφs=0

D. φrᵀφs=0 in all coordinate systems

Correct Answer: C. φrᵀMφs=0

Question 3

What is the main purpose of a modal transformation?

A. To eliminate all damping physically

B. To convert physical coordinates into modal coordinates and often decouple the equations

C. To increase the number of degrees of freedom

D. To impose fixed boundary conditions

Correct Answer: B. To convert physical coordinates into modal coordinates and often decouple the equations

Question 4

A mode is weakly excited by a force applied at a coordinate where its mode-shape amplitude is very small. This is primarily due to low:

A. Modal participation

B. Mass normalization

C. Wave speed

D. Static stiffness

Correct Answer: A. Modal participation

Question 5

Which damping model generally preserves real modal decoupling?

A. C=αM+βK

B. C=M⁻¹K

C. C=αK⁻¹

D. C=0 for every physical system

Correct Answer: A. C=αM+βK

Question 6

In a free-free shaft, a zero-frequency mode generally represents:

A. Material failure

B. A rigid-body motion

C. The highest bending mode

D. Numerical instability only

Correct Answer: B. A rigid-body motion

Question 7

Which theory is generally more suitable than Euler–Bernoulli theory for short, thick beams at higher frequencies?

A. Rigid-body theory

B. Timoshenko beam theory

C. Thin-plate Kirchhoff theory only

D. Quasi-static theory

Correct Answer: B. Timoshenko beam theory

Question 8

Why can a response sensor near a modal node record a small vibration while another sensor records a large vibration at the same frequency?

A. The structure has no natural frequency

B. The mode shape has spatially varying amplitudes

C. Damping is always zero at nodes

D. The mass matrix becomes singular at resonance

Correct Answer: B. The mode shape has spatially varying amplitudes

Question 9

What is a major risk of retaining only low-frequency modes in a reduced model?

A. It always creates extra physical modes

B. It may omit local flexibility and high-frequency or interface effects

C. It makes all natural frequencies zero

D. It prevents calculation of static displacement in every case

Correct Answer: B. It may omit local flexibility and high-frequency or interface effects

Question 10

In experimental modal analysis, why is frequency agreement alone insufficient for model validation?

A. Natural frequencies cannot be measured

B. Different mode shapes can have similar frequencies

C. Mode shapes do not depend on boundary conditions

D. Frequency-response functions contain no phase information

Correct Answer: B. Different mode shapes can have similar frequencies